1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 1995 Linus Torvalds
5 * This file contains the setup_arch() code, which handles the architecture-dependent
6 * parts of early kernel initialization.
8 #include <linux/acpi.h>
9 #include <linux/console.h>
10 #include <linux/crash_dump.h>
11 #include <linux/dma-map-ops.h>
12 #include <linux/dmi.h>
13 #include <linux/efi.h>
14 #include <linux/ima.h>
15 #include <linux/init_ohci1394_dma.h>
16 #include <linux/initrd.h>
17 #include <linux/iscsi_ibft.h>
18 #include <linux/memblock.h>
19 #include <linux/panic_notifier.h>
20 #include <linux/pci.h>
21 #include <linux/root_dev.h>
22 #include <linux/hugetlb.h>
23 #include <linux/tboot.h>
24 #include <linux/usb/xhci-dbgp.h>
25 #include <linux/static_call.h>
26 #include <linux/swiotlb.h>
27 #include <linux/random.h>
29 #include <uapi/linux/mount.h>
36 #include <asm/bios_ebda.h>
38 #include <asm/cacheinfo.h>
42 #include <asm/hypervisor.h>
43 #include <asm/io_apic.h>
44 #include <asm/kasan.h>
45 #include <asm/kaslr.h>
47 #include <asm/memtype.h>
49 #include <asm/realmode.h>
50 #include <asm/olpc_ofw.h>
51 #include <asm/pci-direct.h>
53 #include <asm/proto.h>
54 #include <asm/thermal.h>
55 #include <asm/unwind.h>
56 #include <asm/vsyscall.h>
57 #include <linux/vmalloc.h>
60 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
61 * max_pfn_mapped: highest directly mapped pfn > 4 GB
63 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
64 * represented by pfn_mapped[].
66 unsigned long max_low_pfn_mapped;
67 unsigned long max_pfn_mapped;
70 RESERVE_BRK(dmi_alloc, 65536);
74 unsigned long _brk_start = (unsigned long)__brk_base;
75 unsigned long _brk_end = (unsigned long)__brk_base;
77 struct boot_params boot_params;
80 * These are the four main kernel memory regions, we put them into
81 * the resource tree so that kdump tools and other debugging tools
85 static struct resource rodata_resource = {
86 .name = "Kernel rodata",
89 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
92 static struct resource data_resource = {
93 .name = "Kernel data",
96 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
99 static struct resource code_resource = {
100 .name = "Kernel code",
103 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
106 static struct resource bss_resource = {
107 .name = "Kernel bss",
110 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
115 /* CPU data as detected by the assembly code in head_32.S */
116 struct cpuinfo_x86 new_cpu_data;
118 struct apm_info apm_info;
119 EXPORT_SYMBOL(apm_info);
121 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
122 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
123 struct ist_info ist_info;
124 EXPORT_SYMBOL(ist_info);
126 struct ist_info ist_info;
131 struct cpuinfo_x86 boot_cpu_data __read_mostly;
132 EXPORT_SYMBOL(boot_cpu_data);
134 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
135 __visible unsigned long mmu_cr4_features __ro_after_init;
137 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
141 static phys_addr_t ima_kexec_buffer_phys;
142 static size_t ima_kexec_buffer_size;
145 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
146 int bootloader_type, bootloader_version;
151 struct screen_info screen_info;
152 EXPORT_SYMBOL(screen_info);
153 struct edid_info edid_info;
154 EXPORT_SYMBOL_GPL(edid_info);
156 extern int root_mountflags;
158 unsigned long saved_video_mode;
160 #define RAMDISK_IMAGE_START_MASK 0x07FF
161 #define RAMDISK_PROMPT_FLAG 0x8000
162 #define RAMDISK_LOAD_FLAG 0x4000
164 static char __initdata command_line[COMMAND_LINE_SIZE];
165 #ifdef CONFIG_CMDLINE_BOOL
166 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
169 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
171 #ifdef CONFIG_EDD_MODULE
175 * copy_edd() - Copy the BIOS EDD information
176 * from boot_params into a safe place.
179 static inline void __init copy_edd(void)
181 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
182 sizeof(edd.mbr_signature));
183 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
184 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
185 edd.edd_info_nr = boot_params.eddbuf_entries;
188 static inline void __init copy_edd(void)
193 void * __init extend_brk(size_t size, size_t align)
195 size_t mask = align - 1;
198 BUG_ON(_brk_start == 0);
199 BUG_ON(align & mask);
201 _brk_end = (_brk_end + mask) & ~mask;
202 BUG_ON((char *)(_brk_end + size) > __brk_limit);
204 ret = (void *)_brk_end;
207 memset(ret, 0, size);
213 static void __init cleanup_highmap(void)
218 static void __init reserve_brk(void)
220 if (_brk_end > _brk_start)
221 memblock_reserve(__pa_symbol(_brk_start),
222 _brk_end - _brk_start);
224 /* Mark brk area as locked down and no longer taking any
229 u64 relocated_ramdisk;
231 #ifdef CONFIG_BLK_DEV_INITRD
233 static u64 __init get_ramdisk_image(void)
235 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
237 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
239 if (ramdisk_image == 0)
240 ramdisk_image = phys_initrd_start;
242 return ramdisk_image;
244 static u64 __init get_ramdisk_size(void)
246 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
248 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
250 if (ramdisk_size == 0)
251 ramdisk_size = phys_initrd_size;
256 static void __init relocate_initrd(void)
258 /* Assume only end is not page aligned */
259 u64 ramdisk_image = get_ramdisk_image();
260 u64 ramdisk_size = get_ramdisk_size();
261 u64 area_size = PAGE_ALIGN(ramdisk_size);
263 /* We need to move the initrd down into directly mapped mem */
264 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
265 PFN_PHYS(max_pfn_mapped));
266 if (!relocated_ramdisk)
267 panic("Cannot find place for new RAMDISK of size %lld\n",
270 initrd_start = relocated_ramdisk + PAGE_OFFSET;
271 initrd_end = initrd_start + ramdisk_size;
272 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
273 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
275 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
277 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
278 " [mem %#010llx-%#010llx]\n",
279 ramdisk_image, ramdisk_image + ramdisk_size - 1,
280 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
283 static void __init early_reserve_initrd(void)
285 /* Assume only end is not page aligned */
286 u64 ramdisk_image = get_ramdisk_image();
287 u64 ramdisk_size = get_ramdisk_size();
288 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
290 if (!boot_params.hdr.type_of_loader ||
291 !ramdisk_image || !ramdisk_size)
292 return; /* No initrd provided by bootloader */
294 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
297 static void __init reserve_initrd(void)
299 /* Assume only end is not page aligned */
300 u64 ramdisk_image = get_ramdisk_image();
301 u64 ramdisk_size = get_ramdisk_size();
302 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
304 if (!boot_params.hdr.type_of_loader ||
305 !ramdisk_image || !ramdisk_size)
306 return; /* No initrd provided by bootloader */
310 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
313 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
314 PFN_DOWN(ramdisk_end))) {
315 /* All are mapped, easy case */
316 initrd_start = ramdisk_image + PAGE_OFFSET;
317 initrd_end = initrd_start + ramdisk_size;
323 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
327 static void __init early_reserve_initrd(void)
330 static void __init reserve_initrd(void)
333 #endif /* CONFIG_BLK_DEV_INITRD */
335 static void __init add_early_ima_buffer(u64 phys_addr)
338 struct ima_setup_data *data;
340 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
342 pr_warn("setup: failed to memremap ima_setup_data entry\n");
347 memblock_reserve(data->addr, data->size);
348 ima_kexec_buffer_phys = data->addr;
349 ima_kexec_buffer_size = data->size;
352 early_memunmap(data, sizeof(*data));
354 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
358 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
359 int __init ima_free_kexec_buffer(void)
361 if (!ima_kexec_buffer_size)
364 memblock_free_late(ima_kexec_buffer_phys,
365 ima_kexec_buffer_size);
367 ima_kexec_buffer_phys = 0;
368 ima_kexec_buffer_size = 0;
373 int __init ima_get_kexec_buffer(void **addr, size_t *size)
375 if (!ima_kexec_buffer_size)
378 *addr = __va(ima_kexec_buffer_phys);
379 *size = ima_kexec_buffer_size;
385 static void __init parse_setup_data(void)
387 struct setup_data *data;
388 u64 pa_data, pa_next;
390 pa_data = boot_params.hdr.setup_data;
392 u32 data_len, data_type;
394 data = early_memremap(pa_data, sizeof(*data));
395 data_len = data->len + sizeof(struct setup_data);
396 data_type = data->type;
397 pa_next = data->next;
398 early_memunmap(data, sizeof(*data));
402 e820__memory_setup_extended(pa_data, data_len);
408 parse_efi_setup(pa_data, data_len);
411 add_early_ima_buffer(pa_data);
414 data = early_memremap(pa_data, data_len);
415 add_bootloader_randomness(data->data, data->len);
416 /* Zero seed for forward secrecy. */
417 memzero_explicit(data->data, data->len);
418 /* Zero length in case we find ourselves back here by accident. */
419 memzero_explicit(&data->len, sizeof(data->len));
420 early_memunmap(data, data_len);
429 static void __init memblock_x86_reserve_range_setup_data(void)
431 struct setup_indirect *indirect;
432 struct setup_data *data;
433 u64 pa_data, pa_next;
436 pa_data = boot_params.hdr.setup_data;
438 data = early_memremap(pa_data, sizeof(*data));
440 pr_warn("setup: failed to memremap setup_data entry\n");
445 pa_next = data->next;
447 memblock_reserve(pa_data, sizeof(*data) + data->len);
449 if (data->type == SETUP_INDIRECT) {
451 early_memunmap(data, sizeof(*data));
452 data = early_memremap(pa_data, len);
454 pr_warn("setup: failed to memremap indirect setup_data\n");
458 indirect = (struct setup_indirect *)data->data;
460 if (indirect->type != SETUP_INDIRECT)
461 memblock_reserve(indirect->addr, indirect->len);
465 early_memunmap(data, len);
470 * --------- Crashkernel reservation ------------------------------
473 /* 16M alignment for crash kernel regions */
474 #define CRASH_ALIGN SZ_16M
477 * Keep the crash kernel below this limit.
479 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
480 * due to mapping restrictions.
482 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
483 * the upper limit of system RAM in 4-level paging mode. Since the kdump
484 * jump could be from 5-level paging to 4-level paging, the jump will fail if
485 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
486 * no good way to detect the paging mode of the target kernel which will be
487 * loaded for dumping.
490 # define CRASH_ADDR_LOW_MAX SZ_512M
491 # define CRASH_ADDR_HIGH_MAX SZ_512M
493 # define CRASH_ADDR_LOW_MAX SZ_4G
494 # define CRASH_ADDR_HIGH_MAX SZ_64T
497 static int __init reserve_crashkernel_low(void)
500 unsigned long long base, low_base = 0, low_size = 0;
501 unsigned long low_mem_limit;
504 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
506 /* crashkernel=Y,low */
507 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
510 * two parts from kernel/dma/swiotlb.c:
511 * -swiotlb size: user-specified with swiotlb= or default.
513 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
514 * to 8M for other buffers that may need to stay low too. Also
515 * make sure we allocate enough extra low memory so that we
516 * don't run out of DMA buffers for 32-bit devices.
518 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
520 /* passed with crashkernel=0,low ? */
525 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
527 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
528 (unsigned long)(low_size >> 20));
532 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
533 (unsigned long)(low_size >> 20),
534 (unsigned long)(low_base >> 20),
535 (unsigned long)(low_mem_limit >> 20));
537 crashk_low_res.start = low_base;
538 crashk_low_res.end = low_base + low_size - 1;
539 insert_resource(&iomem_resource, &crashk_low_res);
544 static void __init reserve_crashkernel(void)
546 unsigned long long crash_size, crash_base, total_mem;
550 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
553 total_mem = memblock_phys_mem_size();
556 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
557 if (ret != 0 || crash_size <= 0) {
558 /* crashkernel=X,high */
559 ret = parse_crashkernel_high(boot_command_line, total_mem,
560 &crash_size, &crash_base);
561 if (ret != 0 || crash_size <= 0)
566 if (xen_pv_domain()) {
567 pr_info("Ignoring crashkernel for a Xen PV domain\n");
571 /* 0 means: find the address automatically */
574 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
575 * crashkernel=x,high reserves memory over 4G, also allocates
576 * 256M extra low memory for DMA buffers and swiotlb.
577 * But the extra memory is not required for all machines.
578 * So try low memory first and fall back to high memory
579 * unless "crashkernel=size[KMG],high" is specified.
582 crash_base = memblock_phys_alloc_range(crash_size,
583 CRASH_ALIGN, CRASH_ALIGN,
586 crash_base = memblock_phys_alloc_range(crash_size,
587 CRASH_ALIGN, CRASH_ALIGN,
588 CRASH_ADDR_HIGH_MAX);
590 pr_info("crashkernel reservation failed - No suitable area found.\n");
594 unsigned long long start;
596 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
597 crash_base + crash_size);
598 if (start != crash_base) {
599 pr_info("crashkernel reservation failed - memory is in use.\n");
604 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
605 memblock_phys_free(crash_base, crash_size);
609 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
610 (unsigned long)(crash_size >> 20),
611 (unsigned long)(crash_base >> 20),
612 (unsigned long)(total_mem >> 20));
614 crashk_res.start = crash_base;
615 crashk_res.end = crash_base + crash_size - 1;
616 insert_resource(&iomem_resource, &crashk_res);
619 static struct resource standard_io_resources[] = {
620 { .name = "dma1", .start = 0x00, .end = 0x1f,
621 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
622 { .name = "pic1", .start = 0x20, .end = 0x21,
623 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
624 { .name = "timer0", .start = 0x40, .end = 0x43,
625 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
626 { .name = "timer1", .start = 0x50, .end = 0x53,
627 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
628 { .name = "keyboard", .start = 0x60, .end = 0x60,
629 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
630 { .name = "keyboard", .start = 0x64, .end = 0x64,
631 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
632 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
633 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
634 { .name = "pic2", .start = 0xa0, .end = 0xa1,
635 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
636 { .name = "dma2", .start = 0xc0, .end = 0xdf,
637 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
638 { .name = "fpu", .start = 0xf0, .end = 0xff,
639 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
642 void __init reserve_standard_io_resources(void)
646 /* request I/O space for devices used on all i[345]86 PCs */
647 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
648 request_resource(&ioport_resource, &standard_io_resources[i]);
652 static bool __init snb_gfx_workaround_needed(void)
657 static const __initconst u16 snb_ids[] = {
667 /* Assume no if something weird is going on with PCI */
668 if (!early_pci_allowed())
671 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
672 if (vendor != 0x8086)
675 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
676 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
677 if (devid == snb_ids[i])
685 * Sandy Bridge graphics has trouble with certain ranges, exclude
686 * them from allocation.
688 static void __init trim_snb_memory(void)
690 static const __initconst unsigned long bad_pages[] = {
699 if (!snb_gfx_workaround_needed())
702 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
705 * SandyBridge integrated graphics devices have a bug that prevents
706 * them from accessing certain memory ranges, namely anything below
707 * 1M and in the pages listed in bad_pages[] above.
709 * To avoid these pages being ever accessed by SNB gfx devices reserve
710 * bad_pages that have not already been reserved at boot time.
711 * All memory below the 1 MB mark is anyway reserved later during
712 * setup_arch(), so there is no need to reserve it here.
715 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
716 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
717 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
722 static void __init trim_bios_range(void)
725 * A special case is the first 4Kb of memory;
726 * This is a BIOS owned area, not kernel ram, but generally
727 * not listed as such in the E820 table.
729 * This typically reserves additional memory (64KiB by default)
730 * since some BIOSes are known to corrupt low memory. See the
731 * Kconfig help text for X86_RESERVE_LOW.
733 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
736 * special case: Some BIOSes report the PC BIOS
737 * area (640Kb -> 1Mb) as RAM even though it is not.
740 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
742 e820__update_table(e820_table);
745 /* called before trim_bios_range() to spare extra sanitize */
746 static void __init e820_add_kernel_range(void)
748 u64 start = __pa_symbol(_text);
749 u64 size = __pa_symbol(_end) - start;
752 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
753 * attempt to fix it by adding the range. We may have a confused BIOS,
754 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
755 * exclude kernel range. If we really are running on top non-RAM,
756 * we will crash later anyways.
758 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
761 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
762 e820__range_remove(start, size, E820_TYPE_RAM, 0);
763 e820__range_add(start, size, E820_TYPE_RAM);
766 static void __init early_reserve_memory(void)
769 * Reserve the memory occupied by the kernel between _text and
770 * __end_of_kernel_reserve symbols. Any kernel sections after the
771 * __end_of_kernel_reserve symbol must be explicitly reserved with a
772 * separate memblock_reserve() or they will be discarded.
774 memblock_reserve(__pa_symbol(_text),
775 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
778 * The first 4Kb of memory is a BIOS owned area, but generally it is
779 * not listed as such in the E820 table.
781 * Reserve the first 64K of memory since some BIOSes are known to
782 * corrupt low memory. After the real mode trampoline is allocated the
783 * rest of the memory below 640k is reserved.
785 * In addition, make sure page 0 is always reserved because on
786 * systems with L1TF its contents can be leaked to user processes.
788 memblock_reserve(0, SZ_64K);
790 early_reserve_initrd();
792 memblock_x86_reserve_range_setup_data();
794 reserve_bios_regions();
799 * Dump out kernel offset information on panic.
802 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
804 if (kaslr_enabled()) {
805 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
811 pr_emerg("Kernel Offset: disabled\n");
817 void x86_configure_nx(void)
819 if (boot_cpu_has(X86_FEATURE_NX))
820 __supported_pte_mask |= _PAGE_NX;
822 __supported_pte_mask &= ~_PAGE_NX;
825 static void __init x86_report_nx(void)
827 if (!boot_cpu_has(X86_FEATURE_NX)) {
828 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
829 "missing in CPU!\n");
831 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
832 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
834 /* 32bit non-PAE kernel, NX cannot be used */
835 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
836 "cannot be enabled: non-PAE kernel!\n");
842 * Determine if we were loaded by an EFI loader. If so, then we have also been
843 * passed the efi memmap, systab, etc., so we should use these data structures
844 * for initialization. Note, the efi init code path is determined by the
845 * global efi_enabled. This allows the same kernel image to be used on existing
846 * systems (with a traditional BIOS) as well as on EFI systems.
849 * setup_arch - architecture-specific boot-time initializations
851 * Note: On x86_64, fixmaps are ready for use even before this is called.
854 void __init setup_arch(char **cmdline_p)
857 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
860 * copy kernel address range established so far and switch
861 * to the proper swapper page table
863 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
864 initial_page_table + KERNEL_PGD_BOUNDARY,
867 load_cr3(swapper_pg_dir);
869 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
870 * a cr3 based tlb flush, so the following __flush_tlb_all()
871 * will not flush anything because the CPU quirk which clears
872 * X86_FEATURE_PGE has not been invoked yet. Though due to the
873 * load_cr3() above the TLB has been flushed already. The
874 * quirk is invoked before subsequent calls to __flush_tlb_all()
875 * so proper operation is guaranteed.
879 printk(KERN_INFO "Command line: %s\n", boot_command_line);
880 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
884 * If we have OLPC OFW, we might end up relocating the fixmap due to
885 * reserve_top(), so do this before touching the ioremap area.
889 idt_setup_early_traps();
893 early_ioremap_init();
895 setup_olpc_ofw_pgd();
897 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
898 screen_info = boot_params.screen_info;
899 edid_info = boot_params.edid_info;
901 apm_info.bios = boot_params.apm_bios_info;
902 ist_info = boot_params.ist_info;
904 saved_video_mode = boot_params.hdr.vid_mode;
905 bootloader_type = boot_params.hdr.type_of_loader;
906 if ((bootloader_type >> 4) == 0xe) {
907 bootloader_type &= 0xf;
908 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
910 bootloader_version = bootloader_type & 0xf;
911 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
913 #ifdef CONFIG_BLK_DEV_RAM
914 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
917 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
918 EFI32_LOADER_SIGNATURE, 4)) {
919 set_bit(EFI_BOOT, &efi.flags);
920 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
921 EFI64_LOADER_SIGNATURE, 4)) {
922 set_bit(EFI_BOOT, &efi.flags);
923 set_bit(EFI_64BIT, &efi.flags);
927 x86_init.oem.arch_setup();
930 * Do some memory reservations *before* memory is added to memblock, so
931 * memblock allocations won't overwrite it.
933 * After this point, everything still needed from the boot loader or
934 * firmware or kernel text should be early reserved or marked not RAM in
935 * e820. All other memory is free game.
937 * This call needs to happen before e820__memory_setup() which calls the
938 * xen_memory_setup() on Xen dom0 which relies on the fact that those
939 * early reservations have happened already.
941 early_reserve_memory();
943 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
944 e820__memory_setup();
949 if (!boot_params.hdr.root_flags)
950 root_mountflags &= ~MS_RDONLY;
951 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
953 code_resource.start = __pa_symbol(_text);
954 code_resource.end = __pa_symbol(_etext)-1;
955 rodata_resource.start = __pa_symbol(__start_rodata);
956 rodata_resource.end = __pa_symbol(__end_rodata)-1;
957 data_resource.start = __pa_symbol(_sdata);
958 data_resource.end = __pa_symbol(_edata)-1;
959 bss_resource.start = __pa_symbol(__bss_start);
960 bss_resource.end = __pa_symbol(__bss_stop)-1;
962 #ifdef CONFIG_CMDLINE_BOOL
963 #ifdef CONFIG_CMDLINE_OVERRIDE
964 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
966 if (builtin_cmdline[0]) {
967 /* append boot loader cmdline to builtin */
968 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
969 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
970 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
975 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
976 *cmdline_p = command_line;
979 * x86_configure_nx() is called before parse_early_param() to detect
980 * whether hardware doesn't support NX (so that the early EHCI debug
981 * console setup can safely call set_fixmap()).
987 if (efi_enabled(EFI_BOOT))
988 efi_memblock_x86_reserve_range();
990 #ifdef CONFIG_MEMORY_HOTPLUG
992 * Memory used by the kernel cannot be hot-removed because Linux
993 * cannot migrate the kernel pages. When memory hotplug is
994 * enabled, we should prevent memblock from allocating memory
997 * ACPI SRAT records all hotpluggable memory ranges. But before
998 * SRAT is parsed, we don't know about it.
1000 * The kernel image is loaded into memory at very early time. We
1001 * cannot prevent this anyway. So on NUMA system, we set any
1002 * node the kernel resides in as un-hotpluggable.
1004 * Since on modern servers, one node could have double-digit
1005 * gigabytes memory, we can assume the memory around the kernel
1006 * image is also un-hotpluggable. So before SRAT is parsed, just
1007 * allocate memory near the kernel image to try the best to keep
1008 * the kernel away from hotpluggable memory.
1010 if (movable_node_is_enabled())
1011 memblock_set_bottom_up(true);
1016 apic_setup_apic_calls();
1018 if (acpi_mps_check()) {
1019 #ifdef CONFIG_X86_LOCAL_APIC
1020 apic_is_disabled = true;
1022 setup_clear_cpu_cap(X86_FEATURE_APIC);
1025 e820__reserve_setup_data();
1026 e820__finish_early_params();
1028 if (efi_enabled(EFI_BOOT))
1031 reserve_ibft_region();
1035 * VMware detection requires dmi to be available, so this
1036 * needs to be done after dmi_setup(), for the boot CPU.
1037 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
1038 * called before cache_bp_init() for setting up MTRR state.
1040 init_hypervisor_platform();
1043 x86_init.resources.probe_roms();
1045 /* after parse_early_param, so could debug it */
1046 insert_resource(&iomem_resource, &code_resource);
1047 insert_resource(&iomem_resource, &rodata_resource);
1048 insert_resource(&iomem_resource, &data_resource);
1049 insert_resource(&iomem_resource, &bss_resource);
1051 e820_add_kernel_range();
1053 #ifdef CONFIG_X86_32
1054 if (ppro_with_ram_bug()) {
1055 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
1056 E820_TYPE_RESERVED);
1057 e820__update_table(e820_table);
1058 printk(KERN_INFO "fixed physical RAM map:\n");
1059 e820__print_table("bad_ppro");
1062 early_gart_iommu_check();
1066 * partially used pages are not usable - thus
1067 * we are rounding upwards:
1069 max_pfn = e820__end_of_ram_pfn();
1071 /* update e820 for memory not covered by WB MTRRs */
1073 if (mtrr_trim_uncached_memory(max_pfn))
1074 max_pfn = e820__end_of_ram_pfn();
1076 max_possible_pfn = max_pfn;
1079 * Define random base addresses for memory sections after max_pfn is
1080 * defined and before each memory section base is used.
1082 kernel_randomize_memory();
1084 #ifdef CONFIG_X86_32
1085 /* max_low_pfn get updated here */
1086 find_low_pfn_range();
1090 /* How many end-of-memory variables you have, grandma! */
1091 /* need this before calling reserve_initrd */
1092 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1093 max_low_pfn = e820__end_of_low_ram_pfn();
1095 max_low_pfn = max_pfn;
1097 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1101 * Find and reserve possible boot-time SMP configuration:
1105 early_alloc_pgt_buf();
1108 * Need to conclude brk, before e820__memblock_setup()
1109 * it could use memblock_find_in_range, could overlap with
1116 memblock_set_current_limit(ISA_END_ADDRESS);
1117 e820__memblock_setup();
1120 * Needs to run after memblock setup because it needs the physical
1128 efi_mokvar_table_init();
1131 * The EFI specification says that boot service code won't be
1132 * called after ExitBootServices(). This is, in fact, a lie.
1134 efi_reserve_boot_services();
1136 /* preallocate 4k for mptable mpc */
1137 e820__memblock_alloc_reserved_mpc_new();
1139 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1140 setup_bios_corruption_check();
1143 #ifdef CONFIG_X86_32
1144 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1145 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1149 * Find free memory for the real mode trampoline and place it there. If
1150 * there is not enough free memory under 1M, on EFI-enabled systems
1151 * there will be additional attempt to reclaim the memory for the real
1152 * mode trampoline at efi_free_boot_services().
1154 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1155 * are known to corrupt low memory and several hundred kilobytes are not
1156 * worth complex detection what memory gets clobbered. Windows does the
1157 * same thing for very similar reasons.
1159 * Moreover, on machines with SandyBridge graphics or in setups that use
1160 * crashkernel the entire 1M is reserved anyway.
1162 x86_platform.realmode_reserve();
1166 idt_setup_early_pf();
1169 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1170 * with the current CR4 value. This may not be necessary, but
1171 * auditing all the early-boot CR4 manipulation would be needed to
1174 * Mask off features that don't work outside long mode (just
1177 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1179 memblock_set_current_limit(get_max_mapped());
1182 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1185 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1186 if (init_ohci1394_dma_early)
1187 init_ohci1394_dma_on_all_controllers();
1189 /* Allocate bigger log buffer */
1192 if (efi_enabled(EFI_BOOT)) {
1193 switch (boot_params.secure_boot) {
1194 case efi_secureboot_mode_disabled:
1195 pr_info("Secure boot disabled\n");
1197 case efi_secureboot_mode_enabled:
1198 pr_info("Secure boot enabled\n");
1201 pr_info("Secure boot could not be determined\n");
1208 acpi_table_upgrade();
1209 /* Look for ACPI tables and reserve memory occupied by them. */
1210 acpi_boot_table_init();
1216 early_platform_quirks();
1218 early_acpi_boot_init();
1221 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1223 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1224 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1227 * Reserve memory for crash kernel after SRAT is parsed so that it
1228 * won't consume hotpluggable memory.
1230 reserve_crashkernel();
1232 memblock_find_dma_reserve();
1234 if (!early_xdbc_setup_hardware())
1235 early_xdbc_register_console();
1237 x86_init.paging.pagetable_init();
1242 * Sync back kernel address range.
1244 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1247 sync_initial_page_table();
1253 x86_32_probe_apic();
1258 * Read APIC and some other early information from ACPI tables.
1264 * get boot-time SMP configuration:
1269 * Systems w/o ACPI and mptables might not have it mapped the local
1270 * APIC yet, but prefill_possible_map() might need to access it.
1272 init_apic_mappings();
1274 prefill_possible_map();
1279 io_apic_init_mappings();
1281 x86_init.hyper.guest_late_init();
1283 e820__reserve_resources();
1284 e820__register_nosave_regions(max_pfn);
1286 x86_init.resources.reserve_resources();
1288 e820__setup_pci_gap();
1291 #if defined(CONFIG_VGA_CONSOLE)
1292 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1293 conswitchp = &vga_con;
1296 x86_init.oem.banner();
1298 x86_init.timers.wallclock_init();
1301 * This needs to run before setup_local_APIC() which soft-disables the
1302 * local APIC temporarily and that masks the thermal LVT interrupt,
1303 * leading to softlockups on machines which have configured SMI
1304 * interrupt delivery.
1310 register_refined_jiffies(CLOCK_TICK_RATE);
1313 if (efi_enabled(EFI_BOOT))
1314 efi_apply_memmap_quirks();
1320 #ifdef CONFIG_X86_32
1322 static struct resource video_ram_resource = {
1323 .name = "Video RAM area",
1326 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1329 void __init i386_reserve_resources(void)
1331 request_resource(&iomem_resource, &video_ram_resource);
1332 reserve_standard_io_resources();
1335 #endif /* CONFIG_X86_32 */
1337 static struct notifier_block kernel_offset_notifier = {
1338 .notifier_call = dump_kernel_offset
1341 static int __init register_kernel_offset_dumper(void)
1343 atomic_notifier_chain_register(&panic_notifier_list,
1344 &kernel_offset_notifier);
1347 __initcall(register_kernel_offset_dumper);